Identity Authentication Based On Computer Keystroke Dynamics

In order to protect computer system, password is the most widely used method to control the access of computer system by authenticating the users' identity. However, the policy of password is vulnerable to impostor attacks due to its simplicity. Therefore, the original intention of keystroke dynamics is to provide a convenient and effective protection on password verification. In keystroke dynamics, the characteristics of a user' keystrokes is captured when password is being typed, and is analyzed by algorithms to automatically recognize the authentic identity. What is critical for such an auxiliary identity authentication is to find a recognition algorithm with high accuracy and high speed. Up to now the known recognition algorithms mainly come from the fields of statistics, neural networks, fuzzy logic, etc, but improvements are required because satisfactory accuracy and speed cannot be achieved simultaneously, and most of them are not suitable in detecting novel impostors. In this thesis, these problems are explored in depth, and an identity authentication system based on computer keystroke dynamics is designed and implemented.The thesis firstly presents the overall identity authentication system design, including the levels of data acquisition, data preprocessing, data analysis, and user management. Based on the overall system design, the research and experiment of data acquisition are conducted. Then, Levenberg-Marquardt algorithm is adopted for data analysis. The entire procedures of authentication are simulated in MATLAB and the results are compared with those of previous research both in accuracy and speed. As for the problem of novel impostor detection, the idea of adding equi-distant mean noise sequences into training set is presented. Based on support vector machine, a generalized algorithm is adopted for password verification, together with a comparison between SVM and LM. Finally, some initial works of identity authentication based on mouse trajectories are explored, including the construction of dynamic model, system architecture, data acquisition and analysis, etc.